save the mesh

src/pixel3dmm/tracking/tracker.py

@@ -201,7 +201,7 @@ class Tracker(object):
             self.save_folder = env_paths.TRACKING_OUTPUT
         self.output_folder = os.path.join(self.save_folder, self.actor_name)
         self.checkpoint_folder = os.path.join(self.save_folder, self.actor_name, "checkpoint")
-        self.mesh_folder = os.path.join(self.save_folder, self.actor_name, "mesh")
+        self.mesh_folder = os.path.join(self.save_folder, self.video_name, "mesh")
         self.create_output_folders()
         self.writer = SummaryWriter(log_dir=self.save_folder + self.actor_name + '/logs')
 
@@ -1288,6 +1288,28 @@ class Tracker(object):
                 normals = np.transpose(normals, (1, 2, 0))
                 # scale to uint8
                 arr = (normals * 255).clip(0,255).astype(np.uint8)
+
+                if visualizations == [[View.SHAPE]]:
+                    # build your normal‐map preview as before
+                    normals = ops['normal_images'][0].cpu().numpy()     # [3,H,W]
+                    normals = (normals + 1.0) / 2.0                      # → [0,1]
+                    normals = np.transpose(normals, (1,2,0))            # H×W×3
+                    arr = (normals * 255).clip(0,255).astype(np.uint8)
+
+                    # --- new: export the *non*-canonical mesh ---
+                    os.makedirs(self.mesh_folder, exist_ok=True)
+                    frame_id = str(self.frame).zfill(5)
+                    ply_path = os.path.join(self.mesh_folder, f"{frame_id}.ply")
+
+                    # faces is the same face list you loaded in setup_renderer
+                    faces_np = self.faces.cpu().numpy()  
+                    # vertices is your posed mesh from above:
+                    #    vertices = …  # torch.Tensor of shape (1, V, 3)
+                    verts_np = vertices[0].detach().cpu().numpy()
+
+                    trimesh.Trimesh(faces=faces_np, vertices=verts_np, process=False) \
+                        .export(ply_path)
+
                 return arr
 
             mask = (self.parse_mask(ops, batch, visualization=True) > 0).float()